For modeling thermal heat transfer, not only the effects of convection and conduction are relevant, but also thermal and visible radiation. Radiation is especially important for setups with large temperature differences, as well as for interaction with external light sources.Common computational fluid dynamic models usually treat radiation transport as a minor effect, that can be handled by simplified algorithms. All these normal models, e.g. surface to surface model, discrete transfer model, P_N method, discrete ordinates model, exhibit disadvantages in the computing performance and the physical modeling.
Hence, there are many technical applications, where the fluid simulation are limited both in accuracy and calculation time by the available radiation model. As exemplary cases combustion chambers, smoke and soot creation, solar power generation, UV water disinfection, condensation in car headlights, fusion and fission reactor chambers, electric arc movement, as well as low-emissivity glass windows can be named.
In the fields investigating radiation as main effect, e.g. cinematic 3d animation or illumination simulation for lamps and workspaces, the mentioned methods are not in use anymore as ray tracing is the first choice.
In this work, the existing methods for ray tracing were adapted and implemented with the goal to interact with fluid flow simulations and replace existing radiation modeling. This can be regarded as innovative, interdisciplinary method for the interaction of fluids and solids with radiation, incorporating physical effects that could not be included in previous simulations.
While in usual light calculations, the geometry exists solely in the form of surfaces and their triangulation, fluid flow requires volumetric calculation grids. Hence, methods are implemented that actually use the volumetric grid, and incorporate volumetric effects with little additional effort.
Spectral volumetric path tracing with Monte Carlo integrated, importance sampled emission was hence the method of choice for this work.
The implemented ray tracer is able to emit radiation from point sources, geometric surfaces, as well as from volumetric sources. Spectral dependence of material values is treated using radiation bands with hardly no increase of calculation time, whereas in all other models, the calculation time scales linearly with the amount of bands. Direct, diffuse and mixed surface reflection is modeled. The volumetric refraction index is implemented, so refraction is modeled, even including partial and total reflexion. The focusing of lenses or mirror systems can hence be simulated satisfactory, which cannot be treated sufficiently by any other radiation model. Surface and volumetric absorption are implemented, as well as surface and volumetric scattering effects.
The radiation emission can be caused by a temperature field at surfaces and volumes. These fields are imported from software calculating the fluid and the thermal system. Ray tracing results in volumetric and surface heat sources that can be returned to the original code, and their effect further calculations.
This coupling was implemented and tested with the commercial computational fluid dynamics code Fluent, using its plug-in interface. As most of Fluent's radiation models are only performed after a fixed number of implicit flow and turbulence iterations, no further disadvantages or limitations occur, that are not as well existing for the existing radiation simulations. A fully implicit treatment of radiation is unlikely to be performed, as stability is already sufficient for most applications. Of course, systems containing only heat sources caused by light and no secondary heat radiation can be treated by the implemented ray tracer with high performance.
The implemented ray tracer is validated with analytically solved systems, and compared to quantitative simulation results of other simulation methods. Also, the scattering effects are validated against experimental and simulation results from literature.
The observed calculation performance is similar or faster then for standard models with geometries of approximately 150000 volume elements, while the modeling is done more accurately. For larger models, even larger advantages can be expected.
2010. , 98 p.